torch.eye()函数用来生层对角矩阵，即对角线上的元素都为1，其余都为0.
import torch as t
t.eye(3,3)
# tensor([[1., 0., 0.],
# [0., 1., 0.],
# [0., 0., 1.]])
t.eye(3,4)
# tensor([[1., 0., 0., 0.],
# [0., 1., 0., 0.],
# [0., 0., 1., 0.]])
t.eye(3)
# tensor([[1., 0., 0.],
# [0., 1., 0.],
# [0., 0., 1.]])
squeeze()和unsqueeze()函数squeeze(dim=int) 的用法主要就是对数据的维度进行压缩或者解压, 去掉维数为1的维度。
unsqueeze(dim=int) 的用法是对数据的维度进行扩充,给指定位置加上维数为一的维度。
import torch as t
t = t.rand(1,2,10)
# tensor([[[0.9775, 0.5134, 0.7421, 0.0982, 0.3329, 0.6887, 0.7929, 0.9276, 0.4891, 0.5787],
# [0.1926, 0.7890, 0.4820, 0.0320, 0.1211, 0.7013, 0.8026, 0.7744, 0.5264, 0.4967]]])
t.squeeze(dim=0)
# tensor([[0.9775, 0.5134, 0.7421, 0.0982, 0.3329, 0.6887, 0.7929, 0.9276, 0.4891, 0.5787],
# [0.1926, 0.7890, 0.4820, 0.0320, 0.1211, 0.7013, 0.8026, 0.7744, 0.5264, 0.4967]])
t.unsqueeze(dim=0)
# tensor([[[[0.9775, 0.5134, 0.7421, 0.0982, 0.3329, 0.6887, 0.7929, 0.9276, 0.4891, 0.5787],
# [0.1926, 0.7890, 0.4820, 0.0320, 0.1211, 0.7013, 0.8026, 0.7744, 0.5264, 0.4967]]]])
t.unsqueeze(dim=2)
print(t)
# tensor([[[0.9775, 0.5134, 0.7421, 0.0982, 0.3329, 0.6887, 0.7929, 0.9276, 0.4891, 0.5787],
# [0.1926, 0.7890, 0.4820, 0.0320, 0.1211, 0.7013, 0.8026, 0.7744, 0.5264, 0.4967]]])
# print(t.size())